Exploratory Data Analysis of Time Series Using Pre-segmented Clustering

Time series clustering is an unsupervised method of organizing homogeneous time series in groups based on certain similarity criteria. As a result, it can be an essential step in Exploratory Data Analysis (EDA), especially for complex time series data. This applies specifically to industrial datasets for applications like predictive maintenance, energy consumption, etc., due to the heterogeneity and peculiarity of collected data sets. Understanding the underlying trends and patterns in such datasets could help strategize advanced analysis methods such as forecasting, regression testing, etc. In this paper, we present a case study on a real-world energy consumption dataset of 4G cells, where we perform a pre-segmented clustering based EDA to uncover hidden insights about the data. The empirical study demonstrates that performing pre-segmented clustering based EDA enhances data interpretation by revealing prevalent and infrequent patterns, empowering users to refine analyses such as prediction more precisely, leading to performance improvement.